Privacy Policy

This policy describes how Pulsar67 LLC (“Pulsar,” “we,” “us”) handles information when you use the Pulsar AI chat service. We do not sell personal data or chat content. We do not use your chats to train our AI models, third-party foundation models, or to assist third-party training.

1. How Pulsar works (technical overview)

This section is meant to be readable and technically honest: what is encrypted, what must be plaintext to generate a reply, what we store, and what we cannot see. Engineers should recognize a real architecture — not marketing vapor. Legal terms for use of the Service are in the Terms of Service.

1.1 Account data vs chat content

Account and billing data (email, name, hashed password, plan/subscription status, Stripe identifiers, API key hashes/prefixes, rate-limit counters, and similar operational signals) are stored so we can run accounts, billing, and abuse prevention. That is separate from chat content (message bodies, titles, and related conversation text). We do not treat stored signed-in chat history as readable data we collect for monitoring or training.

1.2 Signed-in chat: client-side encryption at rest

For signed-in users, persisted chat history is encrypted in the browser before it is stored. Staff and database operators cannot read plaintext chat history from our production database.

This is at-rest encryption of stored history. It is not end-to-end encryption of the inference path: prompts must be plaintext to the model while a reply is generated.

1.3 Inference: plaintext for generation

To generate a reply, the model must receive the prompt in plaintext. Pulsar’s models may run on GPU servers we operate and/or on a variety of third-party hosting and infrastructure services we use to serve those models. In either case, the models are Pulsar’s; for the duration of a completion, our application path and the model-serving infrastructure necessarily process the prompt and streaming reply text in plaintext as required to produce a response.

1.4 Guest (signed-out) chat

Guest sessions are limited and temporary. They are not covered by the signed-in password-derived DEK encryption flow. Guest turns may be held briefly in a server-side session cache so the conversation can continue, then expire (on the order of a day). Guest chat does not include Pro features such as reasoning or web search; those are forced off for guest completions. Message count is capped (default on the order of a handful of turns) before signup is required.

1.5 Rate limits and abuse signals

We enforce rate limits and plan entitlements using account- and service-level signals (for example request counters, timestamps, and similar operational logs). Suspension or throttling for abuse is based on how the Service is used — not by reading ciphertext chat histories. Consistent with our privacy design, we cannot review prior encrypted chat content in production.

1.6 Model-embedded safeguards

Safeguards intended to help reduce harmful or prohibited use are built into the models themselves (for example refusal-oriented behavior at generation time). These are not a guarantee of perfect safety, and they are not a substitute for reviewing chat histories — which our at-rest design does not allow staff to do.

2. Information we collect

We do not collect your chat messages, titles, or related chat content as readable data for our own use, monitoring, or training. Section 1 describes how chat storage and encryption work.

3. How we use data

We use the information we collect solely to:

We do not sell personal data or chat content. We do not use your chats to train our AI models, third-party foundation models, or to assist third-party training.

4. Who can see prompts

5. Retention

6. Security

We use industry-standard practices for account authentication, TLS in transit (when HTTPS is enabled), and client-side AES-GCM for chat at rest. No method of transmission or storage is perfectly secure. Plaintext during inference is required to generate replies; at-rest encryption is designed so stored history is not readable chat content we collect or review.

7. Children

Pulsar is not directed to children under 13, and we do not knowingly collect their personal information.

8. Changes

We may update this policy. The “Last updated” date will change when we do. Continued use of the service after an update means you accept the revised policy.

9. Contact

Pulsar67 LLC. Privacy questions or data requests: privacy@example.com